You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Problem — Cross-campaign handoffs between research and dev are vibes: a finding that needs code becomes a prose remark, an observation that deserves an experiment becomes a TODO — nothing typed, nothing evidence-backed, nothing round-trippable. Approach — The handoff seed schemas as committed fixtures in amicode: issue seeds (research → dev) and hypothesis seeds (dev → research), their rendering templates (write-an-issue decision surface; hypothesis card), and the round-trip + adversarial test corpus. Scope — in: seed schemas, both templates, ≥ 4 valid + ≥ 3 adversarial fixtures, the round-trip test · out: the emitting logic (mode cards' closing artifacts), filing flows (existing write-an-issue / vault conventions).
Acceptance Criteria
Issue seed schema: title, motivation, evidence (pointers that must resolve), suggested_repo, suggested_tier — required; kind field fixed to issue
Hypothesis seed schema: observation, evidence (pointers that must resolve), suggested_experiment — required; kind fixed to hypothesis
Round-trip: seed → target artifact body (via the committed template) → re-extract → field-equal over the full declared field set; ≥ 2 valid fixtures per kind
Adversarial refusals (≥ 3, violated schema path named): dead evidence pointers · kind/field-set disagreement (kind issue without suggested_repo; kind hypothesis without suggested_experiment — the masquerade) · missing required field
Zero blocklisted strings
Testing Decisions
New surface; this slice delivers the spec's handoff_roundtrip_pass criterion complete (fixtures + tests together, TDD-shaped).
Key Decisions
Evidence pointers must resolve at validation time — a seed citing a nonexistent artifact is refused, not warned. The masquerade test is mechanical (kind vs field set), not semantic. Filing agency stays protocol discipline (the receiving director or a human files through the normal flows; nothing auto-files) — honestly scoped, enforcement is OB-M5.
Constraints & Invariants
Seeds are the only sanctioned cross-campaign currency beyond issues/vault-cards/pulse-catalog themselves. No mode-private state rides a seed.
Prior Art
write-an-issue's decision-surface template (the issue-seed render target); the vault's hypothesis card convention; the spec's D3.
Source
Part of 497 · spec spec-20260822-063957-auto-rd-two-modes.md (D3, Measurement Protocol) · no blockers.
Important
Problem — Cross-campaign handoffs between research and dev are vibes: a finding that needs code becomes a prose remark, an observation that deserves an experiment becomes a TODO — nothing typed, nothing evidence-backed, nothing round-trippable.
Approach — The handoff seed schemas as committed fixtures in amicode: issue seeds (research → dev) and hypothesis seeds (dev → research), their rendering templates (write-an-issue decision surface; hypothesis card), and the round-trip + adversarial test corpus.
Scope — in: seed schemas, both templates, ≥ 4 valid + ≥ 3 adversarial fixtures, the round-trip test · out: the emitting logic (mode cards' closing artifacts), filing flows (existing write-an-issue / vault conventions).
Acceptance Criteria
issuehypothesisissuewithout suggested_repo; kindhypothesiswithout suggested_experiment — the masquerade) · missing required fieldTesting Decisions
New surface; this slice delivers the spec's
handoff_roundtrip_passcriterion complete (fixtures + tests together, TDD-shaped).Key Decisions
Evidence pointers must resolve at validation time — a seed citing a nonexistent artifact is refused, not warned. The masquerade test is mechanical (kind vs field set), not semantic. Filing agency stays protocol discipline (the receiving director or a human files through the normal flows; nothing auto-files) — honestly scoped, enforcement is OB-M5.
Constraints & Invariants
Seeds are the only sanctioned cross-campaign currency beyond issues/vault-cards/pulse-catalog themselves. No mode-private state rides a seed.
Prior Art
write-an-issue's decision-surface template (the issue-seed render target); the vault's hypothesis card convention; the spec's D3.
Source
Part of 497 · spec
spec-20260822-063957-auto-rd-two-modes.md(D3, Measurement Protocol) · no blockers.